Opalite Health Wants to Retire the 30-Minute Wait for a Medical Interpreter

The interpreter is thirty minutes out. The patient is in pain and speaks Cantonese. The doctor speaks English. Somewhere in that gap - between what a patient can say and what a clinician can understand - care gets slower, thinner, and sometimes wrong. Opalite Health, a Y Combinator Winter 2026 company based in San Francisco, was built to close that gap in the time it takes to say a sentence.

Opalite makes an AI medical interpreter. A provider speaks, a patient speaks back, and the software translates between them in real time across more than 150 languages and dialects - Spanish, Mandarin, Vietnamese, Haitian Creole, Amharic, Swahili, and a long tail of others. It runs 24/7, needs no advance scheduling, and is built to sit inside the tools clinicians already use rather than beside them.

The number that frames the whole company is this: more than 30 million Americans have limited English proficiency. When they see a doctor, the law and good medicine both require an interpreter. The usual options are a human on a phone line or a video cart wheeled into the room, and both come with wait times, scheduling, and a bill that at a large hospital can reach $2 million a year.

There is a quieter cost, too, one that does not show up on an invoice. When an interpreter is not immediately available, providers improvise. They lean on a bilingual family member, a nurse who took Spanish in college, or, worst of all, a translation app never meant for a clinical conversation. Each of those shortcuts introduces the possibility of a small error that changes a plan of care. Opalite's founders talk about the problem less as a translation gap and more as a safety gap that happens to look like a translation gap.

The founders came at this from opposite ends

Cathleen Kuo, the CEO, is a physician and healthcare AI researcher with more than 200 publications to her name. She is also the child of immigrants who do not speak English. She has watched the language barrier from both sides of the exam table - first as the kid translating for her parents, later as the doctor unable to fully reach a patient without help.

Her co-founder and CTO, Alex Mehregan, spent his early career at Apple working on Siri and Apple Intelligence - consumer voice AI used by millions. He is a Berkeley EECS graduate and a second-time founder. The pairing is the pitch: someone who knows exactly where clinical language fails, and someone who knows how to make a voice system work at scale.

That combination shows up in how the product is built. Consumer voice assistants optimize for the average case - the request they hear a thousand times a day. A medical interpreter has to be good at the rare and the consequential: an unusual drug name, a negation ("no chest pain" is the opposite of "chest pain"), a dosage, a hedge in a patient's own words. Opalite describes a pipeline tuned for exactly those failure points, with medical-specialized speech recognition, clinical language handling that expands abbreviations and catches negations, and translation trained on medical language rather than general web text.

What "better than a human" actually means here

Opalite's central claim is uncomfortable and specific: in clinical evaluations, its system produced far fewer errors than certified human interpreters. Physician-researchers - the company cites work involving Johns Hopkins Medicine and the U.S. Department of Veterans Affairs - compared the AI against certified interpreters for Cantonese, Mandarin, and Spanish. The gap was not small.

The reason this matters more in medicine than in a travel app is obvious once you say it out loud. A mistranslated street name is an inconvenience. A mistranslated dosage, allergy, or symptom is a safety event. That is why Opalite built a piece called Guardian - a sentence-level safety layer that scores each interpreted utterance for confidence before it reaches the other person in the room. The point is not just to translate, but to know when it might be wrong.

It is worth being precise about what these numbers are and are not. They come from clinical evaluations rather than a marketing deck, comparing the system against certified interpreters on specific language pairs. They do not mean human interpreters are careless; they mean that a person working quickly, under pressure, across a full day of calls, will miss things that a system checking its own confidence on every sentence can catch. The interesting claim is not "AI is smarter." It is "AI does not get tired, and it can grade itself in real time."

The business is priced against a real benchmark

Opalite charges roughly $2.95 per interpreted minute. The company positions that against the LanguageLine GSA benchmark of about $5.90 a minute - so, about half the cost, or twice the interpreted minutes from the same budget. For a health system already spending heavily on language access under Section 1557 of the Affordable Care Act, that is a line item, not a science project.

There is a second job the software does while it is already in the room: it drafts the visit documentation. Opalite auto-generates clinical notes and audit records, which the company says saves providers about 20% of the time per interpreted session. One tool, two chores - interpret the conversation, then write it up.

That detail is a quiet piece of product strategy. Interpretation on its own is a cost center - something a health system buys because it has to. Documentation is a time drain every clinician already resents. By doing both from the same recording of the same conversation, Opalite turns a compliance expense into something that also gives providers minutes back. The interpreter that also writes your note is a harder thing to rip out than the interpreter that only interprets.

The model, then, is business-to-business software sold by the minute. Opalite's customers are healthcare organizations across categories - hospitals, community health centers, home health agencies, telehealth platforms, and clinics - and it is already used daily by patients and providers in more than ten states. That breadth of customer type matters: a home health nurse visiting a patient's apartment has very different constraints than a hospital front desk, and a system that works in both is harder to build than a demo that works in one.

Where it sits, and what defends it

The incumbents are the language-services giants - LanguageLine, Cyracom, video-cart interpretation - who mostly connect clinicians to human interpreters by phone or screen. Opalite's argument is not that those interpreters are bad people; it is that on-demand human coverage cannot scale to 150 languages at two in the morning, and that a self-checking AI does not get tired on the fortieth call of a shift.

A fair skeptic will note that the raw ingredients of speech AI are increasingly open. What is hard to copy is the rest of it: published clinical validation data, EHR integrations (via FHIR into systems like Epic and Cerner) that take months to land, and the credibility to get a Chief Medical Officer to stake patient safety on your software. Opalite's early customer list - names like CIFC Health, Blythedale Children's Hospital, and Clinic by the Bay, alongside home health and telehealth providers - is the part a weekend prototype cannot reproduce.

At CIFC Health, the reported results were a 50% cut in interpretation costs, 22% less time per interpreted visit, and, per the company, the top choice among both patients and providers. Those are the numbers a health system's finance office and its clinicians can agree on, which is a rarer thing than it sounds.

The regulatory backdrop cuts in Opalite's favor. Section 1557 of the Affordable Care Act obligates health systems that receive federal funding to provide meaningful language access, and enforcement of those rules is not getting looser. For a compliance officer, "we cannot reliably staff a Haitian Creole interpreter at 2am" is a real exposure. A tool that is HIPAA compliant, SOC 2 Type II certified, and marketed as Section 1557 ready reframes that obligation from a recurring scramble into a capability that is simply always on. Regulation, in other words, is not just a cost for Opalite's customers - it is the reason the market exists.

None of this makes the outcome certain. Selling into hospitals is slow, integrations take months, and the moment AI touches a clinical decision, questions about medical-device oversight and liability follow close behind. Opalite is still small, still early, and still proving that its accuracy numbers hold across the messy variety of real patients rather than a controlled evaluation. But the shape of the bet is clear, and it is a disciplined one: pick a problem millions of people live with, measure honestly against the incumbent, and build the boring, defensible parts - validation, integration, trust - that a weekend prototype never can.

The tell in the origin story

Plenty of health-AI companies start with a model and go looking for a problem. Opalite started with a kid in a waiting room, translating for the adults who were supposed to be taking care of her. The product is a straight line from that: make sure no one in a clinic is ever the only person who cannot understand what is being said. Whether Opalite becomes the default interpreter for American healthcare will depend on integrations, regulation, and trust earned one visit at a time. The problem it picked, at least, is not going anywhere.